Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

1 watching

Forks

Latest commit

 

History

26 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation


sunny@bengaluru:~$ whoami

Software Engineer — 4+ years shipping production systems.
Now: SDE 2 @ HackerRank, Platforms team — building agents and MCP.
Before: agentic AI and real-time microservices @ Nivetti Systems.
Formerly founding engineer @ EMO.energy (zero → one).

sunny@bengaluru:~$ cat philosophy.txt

Happy paths are easy. Careers are built on the catch blocks.

🚀 Featured Work

C++20 GGUF SIMD Quantization

LLM inference engine written from scratch in C++20 — no ggml, BLAS or OpenMP. Memory-maps GGUF models from the LLaMA family and runs them on CPU with quantized integer-dot matmuls, a KV cache, hand-written SIMD kernels, a custom thread pool and an OpenAI-compatible server. Every kernel is checked bit-identical against a scalar reference, and every optimization is benchmarked against llama.cpp.

TypeScript Monaco AI Tutor

Where code meets reality: a cloud IDE with real-world debugging scenarios (memory leaks, race conditions, slow queries), an AI Socratic tutor that hints without spoiling, code playback, and glass-box analytics showing companies how candidates think — not just whether tests pass.

Python Claude Multimodal Deterministic

"Can I afford this?" answered properly: pay in full / split / installments / wait, decided by a 90-day balance forecast over real transaction history. The model resolves evidence only — receipt amounts, message amendments — behind a bounded tool loop and strict schemas, while deterministic code owns every money decision. Committed caches reproduce the output bit-identically with zero API calls.

Python Claude Whisper Structured Outputs

Notification triage for a WhatsApp-style multimodal stream → notify / digest / mute. A deterministic 12-trigger risk layer with prompt-injection defense wraps schema-enforced LLM reasoning. Committed caches reproduce the full output bit-identically with zero API calls.

TypeScript Next.js Tool Use

An AI co-pilot that drives the email interface itself — filling the compose form, drafting and refining replies, acting on the mailbox rather than just chatting about it.

💼 Experience

HackerRank
Nov 2026 — present

Software Development Engineer 2

Platforms team — building AI agents and MCP integrations across the HackerRank platform · agentic workflows that connect LLMs to the platform's tools and data

Nivetti Systems
Dec 2023 — Oct 2026

Software Engineer — Applications

Agentic AI systems with LangGraph, LLMs & MCP shipped as product features · optimized real-time microservices for throughput & tail latency under heavy network traffic · resilient APIs and pipelines with Python, Go, FastAPI, Kafka/RabbitMQ · built the switch-management GUI and a compliance-audit tool end-to-end

EMO.energy
Feb — Nov 2023

Software Engineer — Full Stack

Founding engineer — owned architecture and delivery of the first product, SENS, from zero to one · distributed backend on FastAPI, Redis, MongoDB, AWS · cut feature delivery time ~30%

Cognizant
Mar — Aug 2022

Software Engineer — Intern

Modular front-end features + automated test coverage that cut production issues ~20%

🛠️ Stack

Python, Go, C++, TypeScript, JavaScript, React, Next.js, Node.js, FastAPI
PostgreSQL, Redis, MongoDB, Elasticsearch, Kafka, RabbitMQ, AWS, Docker, Git

+ the AI layer:   LangGraph MCP Claude Code Ollama Eval Harnesses

🏅 Certifications & Community

  • AWS Academy — Cloud Architecting · Cloud Foundations
  • Cisco CCNA — Switching, Routing & Wireless Essentials · Introduction to Networks
  • Member, GDG Bangalore (2020 — present) · off-keyboard: cricket 🏏

Open to interesting problems in backend, distributed systems, and applied AI.

Portfolio LinkedIn Email

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors